On 28-mrt-2006, at 17:46, [EMAIL PROTECTED] wrote: > >> ----- Original Message ----- >> From: "Ronald Oussoren" <[EMAIL PROTECTED]> >> To: [EMAIL PROTECTED] >> Subject: Re: [Pythonmac-SIG] Pythonmac-SIG Digest, Vol 35, Issue 51 >> Date: Tue, 28 Mar 2006 09:38:11 +0200 >> >> Sigh, the universal python build already runs on 10.3.9 and >> supports ppc and i386. I don't have a PPC64 box so cannot even try >> to support this. Supporting this in the main build would also >> require more special casing because ppc64 processes can only use a >> small part of the OSX ABI. >> >> Ronald > > Not saying it should, I was just making the point that FAT binaries > can be constructed in more than even two ways. So it seems the > numpy build is treating the 'arch' flags as command line options > that can be logically combined and not as seperate binary build > selectors as it should. Sorry for trying to help. I am approriately > chastised now.
My previous message was a bit too sharp, I hadn't intended to punish anyone :-) The numpy build can be failing for several reasons: * Using a too old build of universal macpython, early builds didn't support building extensions on 10.3 * My patch for distutils is buggy * numpy's custom distutils code is interfering with my changes If you're using Xcode 2.1 or later on OSX 10.4 you can several -arch flags and binaries will be build for all named architectures. This is very convenient, otherwise we'd have to teach distutils about the compile+lipo dance you mentioned and I don't want to perform the necessary surgery on distutils to get it to do that if I don't have to :-) Ronald _______________________________________________ Pythonmac-SIG maillist - Pythonmac-SIG@python.org http://mail.python.org/mailman/listinfo/pythonmac-sig
